This is the ASIC Reliability / Failure-Mode Index: a structured, model-by-model breakdown of what actually fails on Bitcoin ASIC miners and their power supplies, built entirely from public sources and graded for how strong that evidence is. It is the data layer behind the prose — instead of “the S17 is unreliable,” you get the specific component, the reported failure mode, how often it surfaces in public reports, and how many independent sources back it.
Every entry here is reported community experience, not an internal D-Central failure rate. Read the provenance notice below before you cite anything, and see our Mining Data Trust & Methodology page for the full sourcing rules. The whole index is free to download as CSV and JSON under a CC BY 4.0 licence, and each model links straight to the matching D-Central repair service.
Quick answer
D-Central's ASIC Reliability / Failure-Mode Index breaks down what actually fails on 30 Bitcoin ASIC miners and PSUs — 102 component-failure families across hashboards, power domains, fans, control boards and PSUs. Every entry is graded A/B/C for source strength and tagged with how often it is reported in public sources, not a measured failure rate.
The most-reported failure across the whole fleet is daisy-chain death: one dead or cold-solder chip takes the whole hashboard offline. Free to download as CSV and JSON.
The six research lenses
- Forum — Reddit / BitcoinTalk threads (BitcoinTalk and GitHub are the quotable primaries)
- Teardown — YouTube and teardown channels
- Repair cost — repair-cost and service-menu data
- Used market — secondary-market used / refurb prices
- Repair web — shop blogs and support docs
- Open source — open-source GitHub issues
Component buckets
- Hashboard (ASIC + solder)
- Power domain (LDO / filter caps / boost / VRM)
- Control board
- Fan / cooling
- PSU
- Other (cabling, connectors, sensors, firmware, thermal interface, ESD/shipping)
Where a power part lives on the hashboard (LDO, filter caps, boost, level shifters) it is a board repair, not a PSU swap. The standalone wall PSU is a separate bucket.
Data-strength grades
- Grade A Strong — the failure pattern is corroborated across multiple independent public sources and research lenses, with repair-cost and at least directional price data.
- Grade B Moderate — solid failure data, but thinner, conflicting or single-lens supporting evidence on price or secondary detail.
- Grade C Limited — emerging or sparse public data. Flagged honestly and not padded; treat as indicative only.
Per-model failing-component breakdown
Antminer (Bitmain)
Antminer S19 BM1398 Grade A · Strong
Verified spec: 76 chips / 38 domains / 2 chips per domain / 0.36 V core / 14→19 V boost / PIC16F1704 (U3)
Aggregated from 7 community-reported component-failure families across REDDIT, YT, RCOST, RWEB. Data strength: STRONG.
| Component | Failure mode (reported) | Frequency | Sources |
|---|---|---|---|
| Hashboard (ASIC + solder) | One dead chip / cold-or-cracked solder joint disables the whole board (downstream chips vanish) (#1 most-reported ASIC failure family) | frequently reported | 4 REDDIT · YT · RCOST · RWEB |
| Power domain (LDO / caps / boost / VRM) | 0.8V/1.8V filter caps short + per-domain LDO -> domain voltage collapse, '0 chips' (Hashboard-resident power; board repair not PSU swap) | frequently reported | 2 RWEB · YT |
| Power domain (LDO / caps / boost / VRM) | Boost 14V->19V (U9) | occasionally to frequently reported | 2 RWEB · YT |
| Fan / cooling | Cooling fan (175x175x50, 6-pin) -> ERROR_FAN_LOST / 0 RPM | frequently reported | 1 YT |
| Other (cabling, connectors, sensors, firmware, thermal interface) | Flat ribbon / 18-pin signal cable + connector -> board 'not detected' / 'less link'; cheapest most-common not-detected cause (cabling) | frequently reported | 3 RCOST · YT · RWEB |
| Other (cabling, connectors, sensors, firmware, thermal interface) | PIC U3 (PIC16F1704) 'Fail to read PIC' (usually actually a cable/connector fault) (PIC) | occasionally reported | 2 REDDIT · RWEB |
| Other (cabling, connectors, sensors, firmware, thermal interface) | ESD handling damage (7nm, <100V) / shipping damage (ESD/shipping) | occasionally to frequently reported | 2 RWEB · YT |
Notable symptoms: “Chain[x] only find N asic, will power off hash board”; “ASIC75”; “ASICNG:(X)”; “PatternNG”.
Repair-cost signal: Board-level hashboard repair is widely listed around US$100–$175 by third-party shops. D-Central's own tiered pricing runs L1 $95 / L2 $145 / L3 $195 CAD; the repair estimator puts an S19 hashboard at roughly $200–$500 CAD, with a replacement board around $291 CAD. Burn-damaged boards are generally not repairable.
Used-market signal: Public marketplace listings clustered around US$300–$900 (range ~US$166–$900; asicprices.com, 2026-06) — roughly CAD $232–$1,256 as an FX-only estimate. A late-2023 reference near US$3,225 shows how steeply the model has depreciated.
Antminer S19 Pro BM1398 Grade B · Moderate
Verified spec: 114 chips / 38 domains / 3 chips per domain / 0.32 V core / 12.6→20 V boost / PIC (U6) / EEPROM (U10)
Aggregated from 5 community-reported component-failure families across REDDIT, YT, RWEB, RCOST. Data strength: MODERATE.
| Component | Failure mode (reported) | Frequency | Sources |
|---|---|---|---|
| Hashboard (ASIC + solder) | ASIC at chain tail + cold solder joint -> partial chip count (pattern frequent; one BitcoinTalk thread anecdote is single) | frequently reported | 3 REDDIT · YT · RWEB |
| Power domain (LDO / caps / boost / VRM) | Per-domain filter caps / LDO -> 0 chips, domain voltage drop | frequently reported | 3 REDDIT · RWEB · YT |
| Power domain (LDO / caps / boost / VRM) | Boost circuit / domain voltage -> 'get power type version failed' | occasionally reported | 1 YT |
| Hashboard (ASIC + solder) | New-unit early-life chip 'Abnormal' (factory/early-life defects) (Bitmain community) | occasionally reported | 1 RCOST |
| Other (cabling, connectors, sensors, firmware, thermal interface) | Cascading bricking after SD-reflash (healthy boards entered 'abnormal state') (BitcoinTalk 5348903; firmware/cascade) | single anecdotal report | 1 REDDIT |
Notable symptoms: “Chain 1 only find 109 asic” (72 TH vs 110); “signal interrupted at chip 109, chip 110 is bad”; “0 ASIC chip”.
Repair-cost signal: Served under the S19-family board-level menus (third-party board repair parallels the ~US$100–$175 S19 tier; D-Central tiered as S19). Confirm scope after diagnosis.
Used-market signal: Thin and conflicting — do not pin a single number. A late-2023 reference near US$3,734; 2026 used units quoted ~US$2,200–$2,600 with warranty (≈CAD $3,015–$3,560); per-TH math (~US$10–12/TH) implies ~US$1,100–$1,320 for a 110 TH unit (≈CAD $1,500–$1,810). Low confidence.
Antminer S19j Pro / S19j Pro+ BM1362 (S19j Pro+ = BM1362BD) Grade A · Strong
Verified spec: 126 chips / 42 domains / 3 chips per domain / 0.30 V core
Aggregated from 6 community-reported component-failure families across REDDIT, YT, RWEB. Data strength: STRONG.
| Component | Failure mode (reported) | Frequency | Sources |
|---|---|---|---|
| Hashboard (ASIC + solder) | Incomplete chip detection 'Chain[0]: find 0 asic' ('most frequently encountered' per Zeus) (chip corrected to BM1362 vs bible-facts.md summary (VERIFY V1)) | frequently reported | 3 REDDIT · YT · RWEB |
| Hashboard (ASIC + solder) | Poor chip soldering / un-tinned pads (pins look tinned but no contact) | frequently reported | 2 REDDIT · RWEB |
| Power domain (LDO / caps / boost / VRM) | 0.8V/1.2V filter caps short + LDO -> 'ASICNG:(0)', domain voltage loss ('most conditions') (single lens but flagged 'most') | frequently reported | 1 RWEB |
| Power domain (LDO / caps / boost / VRM) | Boost 15V->20V (C915) | occasionally reported | 1 RWEB |
| PSU | PSU (APW12) won't power on / dead fans -> replace PSU first (standalone wall PSU bucket) | frequently reported | 1 YT |
| Other (cabling, connectors, sensors, firmware, thermal interface) | Temp sensors U5/U7/U8/U9 (sensors) | occasionally reported | 1 RWEB |
Notable symptoms: “find 0 asic”; “ASIC113” (114 at 115200 baud, 113 at 12M); “PIC sensor NG”.
Repair-cost signal: Board-level repair listed around US$150–$200 (2–3 day turnaround, burn excluded) by third-party shops; D-Central tiered CAD as S19. A skilled-labour job (BGA rework station, 350–380°C, double-tested) underpins the band.
Used-market signal: S19j Pro used ~US$297–$980 (asicprices.com; clear data-error rows at $5k–$6.4k excluded) ≈ CAD $414–$1,367. Refurb-with-warranty (96 TH) around US$1,500 (≈CAD $2,092). S19j Pro+ trades mostly bundled-with-hosting — no clean standalone figure. Negative day-1 profit at low power cost explains the depressed prices.
Accuracy note: Chip is BM1362 (the BM1398 grouping in quick-reference summaries is imprecise; the detailed hardware reference confirms BM1362). Repair guides had it right.
Antminer S19 XP BM1366 Grade B · Moderate
Verified spec: ~100 GH/s per chip, 21.5 J/TH; 110 chips; domain split documented as “varies”
Aggregated from 4 community-reported component-failure families across YT, RWEB. Data strength: MODERATE.
| Component | Failure mode (reported) | Frequency | Sources |
|---|---|---|---|
| Hashboard (ASIC + solder) | BM1366 chip soldering / micro-shorts (0-hundreds ohm) - guide #1 cause; '0 hashrate'/incomplete detection | frequently reported | 2 YT · RWEB |
| Power domain (LDO / caps / boost / VRM) | LDO (1.2V & 0.8V) - no domain voltage despite 14V input | frequently reported | 2 RWEB · YT |
| Other (cabling, connectors, sensors, firmware, thermal interface) | RX signal trace loss; level-shifter board U10 ('false welding' -> counts in 10/20/30...) (signal/level-shifter; VERIFY Zeus domain layout V4) | frequently reported | 1 RWEB |
| Other (cabling, connectors, sensors, firmware, thermal interface) | U1/U2 signal ICs (wrong power-sequence burnout); crystal Y1; EEPROM U6 | occasionally reported | 1 RWEB |
Notable symptoms: “RI signal disappears at damaged/shorted chip”; integer-multiple chip counts on the pattern test; “EEPRXM NG”.
Repair-cost signal: No dedicated S19 XP menu price isolated — served under D-Central tiered L1/L2/L3 CAD plus general S19-series board repair.
Used-market signal: Refurb band around US$1,699–$1,899 (141 TH, asicprices.com) ≈ CAD $2,370–$2,649; an anecdotal sub-US$900 used band exists, and clearance 141 TH units have appeared near US$649.
Accuracy note: A repair-vendor “11 domains / 10-per-domain” layout is not treated as canonical (110 chips confirmed; domain split documented as “varies”).
Antminer S19k Pro BM1366 Grade B · Moderate
Verified spec: BM1366BS / BM1366BP
Aggregated from 2 community-reported component-failure families across YT, RCOST. Data strength: MODERATE.
| Component | Failure mode (reported) | Frequency | Sources |
|---|---|---|---|
| Fan / cooling | Cooling fan (120mm 4-pin square, shared T21/S21 NOT S19 Pro 6-pin) -> insufficient fan speed | occasionally reported | 1 YT |
| Hashboard (ASIC + solder) | ASIC / LDO / caps - standard chip+LDO scope (BM1366BS/BM1366BP) (thin failure detail; flag) | reported (vendor parts list) | 2 YT · RCOST |
Notable symptoms: Insufficient fan speed; standard chip + LDO scope.
Repair-cost signal: Board repair listed around US$150 base + ~US$50/chip by third-party shops; integrated-heatsink boards only (separate-heatsink boards declined).
Used-market signal: New / clearance around US$319–$379 (110–120 TH) ≈ CAD $445–$529; genuine used below US$319 is plausible but unconfirmed.
Accuracy note: Thin on public failure detail — graded honestly.
Antminer S21 / T21 BM1368 Grade A · Strong
Verified spec: 108 chips / 12 domains / 9 chips per domain / ~1.2 V / boost (U206) ~25 V / 11 level shifters / NO PIC / EEPROM (U6) / 25 MHz crystal (Y1)
Aggregated from 8 community-reported component-failure families across REDDIT, YT, RCOST, RWEB. Data strength: STRONG.
| Component | Failure mode (reported) | Frequency | Sources |
|---|---|---|---|
| Hashboard (ASIC + solder) | ASIC chip (BM1368 PA/PB/AA not interchangeable) - chain reports <108; 'find 82 then 84' -> board off | frequently reported | 4 REDDIT · YT · RCOST · RWEB |
| Hashboard (ASIC + solder) | Thermal-shock solder-joint cracking on power-cycle (hot ~70C -> room cycling cracks marginal joints) - 2025 headline (BitcoinTalk 5539199 mikeywith) | frequently reported | 1 REDDIT |
| Power domain (LDO / caps / boost / VRM) | Op-amp / level-shifter at 2-domain handover (SGM8304 / SN74AUP1T34DCKR U2E) - 'almost always an op-amp/level-shifter issue' (LYS) (flagged dominant) | frequently reported | 1 RWEB |
| Power domain (LDO / caps / boost / VRM) | Per-domain LDO (0.8V core / 1.2V VDDIO); MP2019 high-V LDO U166/U200 domains 11/12; boost SY7304DBC | frequently to occasionally reported | 2 RWEB · YT |
| Other (cabling, connectors, sensors, firmware, thermal interface) | EEPROM GT24C02A U6; temp sensors S75 U5/U7 + 3.3V-rail short cascade (measure 3.3V impedance before power) (EEPROM/sensors) | frequently reported | 1 RWEB |
| Other (cabling, connectors, sensors, firmware, thermal interface) | Factory thermal paste (poor application) -> overheating even on new boards (thermal interface) | frequently reported | 2 RWEB · YT |
| Control board | Control board / wiring mimics hashboard failure (swap 3 wires to isolate) | occasionally reported | 1 REDDIT |
| Fan / cooling | Fans (4-pin, shared T21/S19k Pro) | reported | 1 YT |
Notable symptoms: Chain reports below 108 (“find 82 then 84” → board off); chip counts in multiples of 9; “Pattern NG”; one severe post-cycle case read 107/15/0 across three chains.
Repair-cost signal: Board-level repair listed around US$300 (4–5 day, 30-day warranty) by third-party shops; single BM1368 chip swaps as low as ~US$15–$50. D-Central's estimator puts an S21 hashboard at $300–$600 CAD, per-chip $40–$100, control board $150–$300, diagnostic $75–$150.
Used-market signal: Used around US$748 (188 TH, ~$3.98/TH, mid-2026) ≈ CAD $1,025–$1,043; new ~US$1,699; clearance 151 TH near US$989. T21 trades relative only (~$500–$1k under S21, no clean absolute).
Accuracy note: The S21 has no PIC chip — a common point of confusion with earlier Antminers.
Antminer S21 repair service → Antminer S21 error-code reference
Antminer S21 Pro BM1370 Grade B · Moderate
Verified spec: board layout documented as “varies” (chip confirmed BM1370)
Aggregated from 3 community-reported component-failure families across YT, RWEB. Data strength: MODERATE.
| Component | Failure mode (reported) | Frequency | Sources |
|---|---|---|---|
| Hashboard (ASIC + solder) | BM1370 chip (densely packed) - thermal cycling + power surges -> individual chip failures (board layout 'Varies' in Bible; VERIFY V3) | frequently reported | 2 YT · RWEB |
| Other (cabling, connectors, sensors, firmware, thermal interface) | Heatsink / thermal interface - hot-spot chips from poor contact (thermal interface) | frequently reported | 1 YT |
| Power domain (LDO / caps / boost / VRM) | Shares S21-gen level-shifter / LDO / paste family (inference from architecture) (inference) | reported - emerging | 1 RWEB |
Notable symptoms: Hot-spot chips; chain reports below nameplate.
Repair-cost signal: No dedicated S21 Pro menu price isolated — serviced under the S21 class (~US$300 third-party) plus BM1370 chip swap.
Used-market signal: Used around US$1,199 (234–245 TH) ≈ CAD $1,643–$1,672; some eBay refurb and aggregate figures range much higher (treat with caution); new clusters ~US$1,358–$2,500.
Accuracy note: A repair-vendor “65 ch / 13 dom / 5-per” layout is not treated as canonical (board layout documented as “varies”). Power-domain failure modes are inferred from the S21 generation, flagged as emerging.
Antminer S21 XP BM1370 Grade B · Moderate
Verified spec: 91 chips / 13 domains / 7 chips per domain / ~1.04 V / MP2019 boost (U146/U202) / 12 level shifters (U1–U12)
Aggregated from 3 community-reported component-failure families across YT, RWEB. Data strength: MODERATE.
| Component | Failure mode (reported) | Frequency | Sources |
|---|---|---|---|
| Power domain (LDO / caps / boost / VRM) | Boost ICs on domains 11-12 (MP2019) - domains 11/12 read 0V/low (consistent with Bible architecture) | frequently reported | 2 YT · RWEB |
| Power domain (LDO / caps / boost / VRM) | Voltage regulator / LDO - domain ~0V or <<1.2V | frequently reported (S21-gen) | 2 YT · RWEB |
| Hashboard (ASIC + solder) | BM1370 chip - one case 12 burned chips replaced (REJECT '108 chips/chain 324 total' summary V5) | single anecdotal report | 1 YT |
Notable symptoms: Domain-11/12 boost-fault signature; low domain voltage.
Repair-cost signal: No dedicated S21 XP menu price isolated — S21 class (~US$300 third-party) plus BM1370 chip swap.
Used-market signal: Used around US$3,750 (270 TH, ~$13.89/TH) ≈ CAD $5,231; new ~US$3,899 ≈ CAD $5,439. A “with-firmware $8,500” over-ask listing is excluded as an outlier.
Accuracy note: A “108 chips/chain, 324 total” search summary is an S21 mix-up — the S21 XP is 91 chips per board.
Antminer S21+ BM1370 Grade C · Limited
Verified spec: board layout documented as “varies” (chip confirmed BM1370)
Aggregated from 1 community-reported component-failure family across YT, RWEB. Data strength: THIN.
| Component | Failure mode (reported) | Frequency | Sources |
|---|---|---|---|
| Power domain (LDO / caps / boost / VRM) | Shares S21-gen BM1370 boost / LDO / level-shifter / paste family (little S21+-unique teardown data; flag honestly) | reported - emerging | 2 YT · RWEB |
Notable symptoms: Shares the S21-generation BM1370 boost / LDO / level-shifter / thermal-paste family.
Repair-cost signal: Served under the S21 class (~US$300 third-party).
Used-market signal: New ~US$1,749 (235 TH) ≈ CAD $2,440; a high “$4,659 used” listing is excluded as an above-new outlier (likely hosted / firmware-loaded). New-reference clusters $3,350–$4,030 (early 2026).
Accuracy note: Emerging — little S21+-unique teardown data exists; graded honestly.
Antminer T21 BM1368 Grade A · Strong
Verified spec: shares the S21 108-chip board and 4-pin fan
Aggregated from 1 community-reported component-failure family across REDDIT. Data strength: STRONG (failure via shared thread) / THIN (used price).
| Component | Failure mode (reported) | Frequency | Sources |
|---|---|---|---|
| Hashboard (ASIC + solder) | Thermal-shock solder cracking (covered jointly with S21/T21); no T21-unique teardown beyond shared board (see S21/T21 row) | frequently reported | 1 REDDIT |
Notable symptoms: Thermal-shock solder cracking on power-cycle is the headline failure (shared with the S21).
Repair-cost signal: Served under the S21-class board repair (~US$300 third-party).
Used-market signal: Relative only — roughly $500–$1k under the S21; no clean absolute isolated.
Accuracy note: Failure data is strong via the shared S21/T21 thread; used-price data is limited.
Antminer T21 repair service → Antminer S21/T21 error-code reference
Antminer S17 / S17 Pro / S17+ / T17 BM1393 / BM1397 Grade A · Strong
Verified spec: S17 / T17 = BM1393 (~166 GH/s, 45 J/TH); S17+ / T17+ / S17e / T17e = BM1397 (~200 GH/s, 36 J/TH); 12 domains
Aggregated from 5 community-reported component-failure families across REDDIT, YT, RWEB. Data strength: STRONG.
| Component | Failure mode (reported) | Frequency | Sources |
|---|---|---|---|
| Hashboard (ASIC + solder) | Heatsink-adhesive delamination (copper plating peels off chip) -> loose heatsink shorts adjacent PCB -> cascading hashboard damage. Signature S17 failure (thermal interface) | frequently reported | 3 REDDIT · YT · RWEB |
| Hashboard (ASIC + solder) | ASIC solder-joint cracking from thermal cycling - 'most common S17 failure' | frequently reported | 2 YT · RWEB |
| Power domain (LDO / caps / boost / VRM) | VR MOSFETs (4 per board) - failed MOS kills that board's domains | frequently reported | 2 REDDIT · RWEB |
| Other (cabling, connectors, sensors, firmware, thermal interface) | Thermal paste degradation 'measured in months'; hashboard connectors; BM1397 chip (AD/AG/AH/AI interchangeable; <=5 swaps/board) (third-party '30% failure rate' = attributed estimate, NOT fact) | frequently reported | 2 RWEB · YT |
| PSU | PSU APW9/APW9+ output caps fail under load | frequently reported | 1 YT |
Notable symptoms: “Chain 0 only find 32 asic”; loose-heatsink overheat; shipping-induced solder movement.
Repair-cost signal: Board-level repair listed around US$200 (2–3 day, 15-day warranty, burn excluded) by third-party shops; D-Central's estimator puts an S17 at $150–$350 CAD, firmware reflash $50–$100. The higher labour cost reflects difficulty, not unit value.
Used-market signal: End-of-life — no clean current price isolated (low double-/triple-digit USD expected on 50–73 TH listings).
Accuracy note: A third-party repair summary cites a “30% failure rate” from poor solder. That is a third-party estimate quoted with attribution, not a measured rate and not D-Central- or manufacturer-verified; the “least-reliable line” framing is community opinion, not data.
Antminer L7 BM1489 Grade A · Strong
Verified spec: Scrypt, TSMC 7nm; 120 chips/board × 24 domains × 5-per; 4 chains → 480 chips/unit; dual crystals (Y1 ch1–60 / Y2 ch61–120)
Aggregated from 5 community-reported component-failure families across REDDIT, YT, RWEB. Data strength: STRONG.
| Component | Failure mode (reported) | Frequency | Sources |
|---|---|---|---|
| Hashboard (ASIC + solder) | BM1489 ASIC (thermal burnout / shorted pins) - 'most frequently replaced L7 component'; 'detect 0 chips' / <120 (BM1485 = L3+; do NOT conflate (VERIFY V2)) | frequently reported | 3 REDDIT · YT · RWEB |
| Power domain (LDO / caps / boost / VRM) | LDO chain (SGM2036-ADJ 0.8V; MP2019GN; NCP114ASN180T 1.8V) - 'most common drift point on aged L7 boards' | frequently reported | 2 REDDIT · RWEB |
| Power domain (LDO / caps / boost / VRM) | Domain-switch MOSFET VIS30010 - domain 0V despite 15V input | frequently reported | 1 RWEB |
| Power domain (LDO / caps / boost / VRM) | Level translator SN74AUP1T34DCKR (U2E) burned by wrong power-on order - 'most common bench-error cause' (bench-error, not field failure) | frequently reported | 1 RWEB |
| Other (cabling, connectors, sensors, firmware, thermal interface) | Boost MOSFET TPHR9003NL / MP1517DR; dual crystals Y1/Y2; PIC16F1704 U6; temp sensors S75/NCT218; 0.8/1.8V filter caps (crystals: chips 1-60 or 61-120 fail together) | occasionally to frequently reported | 2 RWEB · REDDIT |
Notable symptoms: “ChipSetting_get_addr_LTC detect 0 chips”; chips 1–60 or 61–120 fail together (crystal); “ASIC NG:N”.
Repair-cost signal: Public board-level repair runs roughly US$300/board or ~US$675/machine (with deposit and storage policies); bulk salvage boards have appeared far cheaper.
Used-market signal: Used around US$249 (9.05 GH/s, ~$30/GH) ≈ CAD $347; used 9050M listings appear (some with residual manufacturer warranty), most without surfaced prices.
Accuracy note: Chip is BM1489 (BM1485 is the older L3+ chip — do not conflate).
Whatsminer (MicroBT)
Whatsminer M30S / M30S+ / M30S++ KF1930 Grade A · Strong
Verified spec: MicroBT KF-series; P21-family WhatsPower PSU. Error-code-centric public data.
Aggregated from 5 community-reported component-failure families across REDDIT, YT, RWEB. Data strength: STRONG.
| Component | Failure mode (reported) | Frequency | Sources |
|---|---|---|---|
| PSU | Power-output protection (errors 233-238, 268; overcurrent/over-temp) - 'by far the most frequently reported Whatsminer errors across all models' (PSU often FINE; check copper-bolt torque first) | frequently reported | 3 REDDIT · YT · RWEB |
| Hashboard (ASIC + solder) | Error 542 'loss balance' - one bad board -> whole unit restart-loops; workaround = remove bad board -> ~66% (BitcoinTalk 5392629) | frequently reported | 1 REDDIT |
| Hashboard (ASIC + solder) | ASIC / hashboard (codes 202/203/205/207/217; 550-552 'bad chips') - located via M30/M50 test fixture | frequently reported | 2 YT · RWEB |
| Other (cabling, connectors, sensors, firmware, thermal interface) | Loose PSU<->miner copper-bar bolts - 233-238 reported but PSU often FINE; LYS: re-torque fixes '~95% of no-output cases' (vendor claim, attribute) (connector; vendor 95% claim attributed not fact) | frequently reported | 1 RWEB |
| Other (cabling, connectors, sensors, firmware, thermal interface) | Temp sensor SM0 (errors 300/530; I2C NAK); thermal interface; EEPROM; voltage-bus loose screws (560-562); fan->ribbon vibration (sensors/EEPROM/fan) | occasionally to frequently reported | 3 REDDIT · YT · RWEB |
Notable symptoms: Error 542 never-ending restart loop; “cannot detect SM0”; RXD/TXD 1.8 V / CLK 0.9 V diagnostic.
Repair-cost signal: Board-level repair listed around US$250 (≤3 chips, 2–3 day; severe burn excluded) by third-party shops; D-Central's estimator puts M20/M30 at $150–$400 CAD. Whatsminer repairs generally run higher than Antminer.
Used-market signal: Weak / wide spread — report as a range. Bulk lots near US$75/unit (88 TH) at one extreme; ~US$299 (112 TH, $3/TH) typical used; retail listings up to the ~CAD $1,000 range. Negative profit at higher power costs.
Accuracy note: A re-torque-fixes-“~95% of no-output cases” figure is a vendor claim, quoted with attribution.
Whatsminer M50 / M50S / M50S+ / M50S++ KF1968 / KF1968E Grade B · Moderate
Verified spec: MicroBT KF-series; P221/P222-family WhatsPower PSU.
Aggregated from 3 community-reported component-failure families across REDDIT, YT, RWEB. Data strength: MODERATE.
| Component | Failure mode (reported) | Frequency | Sources |
|---|---|---|---|
| PSU | WhatsPower P221B/P222B/P222C failure -> won't start / abnormal hashrate; codes 208-275/326/329/8700; wrong-PSU keying -> 201/8700 | frequently reported (dominant M50 family) | 3 REDDIT · YT · RWEB |
| Power domain (LDO / caps / boost / VRM) | Hashboard LDO power (errors 540/541/542) - LDO powering failure on hashboard | frequently reported | 2 YT · RWEB |
| Hashboard (ASIC + solder) | ASIC chip - located via M30/M50/M60 fixture; progressive hashboard loss post-update | frequently to occasionally reported | 2 YT · RWEB |
Notable symptoms: “will report some error when running only one hashboard, but works fine”; error 540-series LDO.
Repair-cost signal: Board-level repair listed around US$350 by third-party shops; D-Central's estimator puts M50/M60 at $250–$550 CAD.
Used-market signal: New figures only — do not invent used. ~US$1,699 new (M50, 110 TH); ~US$699 new (M50S, 126 TH); higher M50S++ via GBP retail.
Whatsminer M53 / M56S KF-series (hydro / immersion) Grade C · Limited
Verified spec: hydro / immersion variants; small installed base, mostly hosted / commercial.
Aggregated from 1 community-reported component-failure family across REDDIT, YT, RWEB. Data strength: THIN.
| Component | Failure mode (reported) | Frequency | Sources |
|---|---|---|---|
| Other (cabling, connectors, sensors, firmware, thermal interface) | Folded into M50/M60 hydro LDO/ASIC + PSU error families; water-cooling adds coldplate / leak / pump surface (small installed base, hosted/commercial; flag honestly) | reported - sparse, emerging | 3 REDDIT · YT · RWEB |
Notable symptoms: Folded into the M50/M60 hydro LDO/ASIC and PSU-error families; water cooling adds coldplate / leak / pump surfaces.
Repair-cost signal: No dedicated public menu isolated — served under general Whatsminer menus plus D-Central tiered.
Used-market signal: M53 = none isolated; M56S+ retail around GBP £1,600 (224 TH, condition unstated) ≈ CAD $2,830 as an FX-only estimate.
Accuracy note: Thin — small installed base; graded honestly, not padded.
Avalon (Canaan)
Avalon A1246 A3210 Grade A · Strong
Verified spec: 38 chips/board × 3 = 114; ~85–90 TH; AUC3 controller. Per-chip BIN matching matters on replacement.
Aggregated from 6 community-reported component-failure families across REDDIT, YT, RWEB. Data strength: STRONG.
| Component | Failure mode (reported) | Frequency | Sources |
|---|---|---|---|
| Hashboard (ASIC + solder) | Whole-board death (RagingTalonS 'hashed 1-2 days then died'; spoonbandits ~8 weeks; happysquid H2=-273C) | frequently reported (early + mid-life) | 3 REDDIT · YT · RWEB |
| Power domain (LDO / caps / boost / VRM) | 17.5V boost module (daughter-card) - 'most distinctive A1246 failure'; replaceable as sub-assembly | frequently reported | 3 REDDIT · YT · RWEB |
| Hashboard (ASIC + solder) | A3210 ASIC (thermal degradation, aged 3-5 yr) - one dead chip kills all downstream (AUC3 estats first 'x' = break) | frequently reported (aging fleet) | 3 REDDIT · YT · RWEB |
| Other (cabling, connectors, sensors, firmware, thermal interface) | AUC3 USB connector corrosion / vibration drops - single most-reported comms-error cause in Avalon threads (connector) | frequently reported | 3 REDDIT · YT · RWEB |
| Other (cabling, connectors, sensors, firmware, thermal interface) | AP1084 LDO; crystal 813RN (25 MHz); MAX14930 isolator; burned MOSFET drivers; output filter caps; 0ohm jumpers; dried paste | occasionally to frequently reported | 2 RWEB · YT |
| Other (cabling, connectors, sensors, firmware, thermal interface) | Temp-sensor misread (~120-180C); fan stuck ~100% RPM (sensor/fan) | single anecdotal report (each) | 1 REDDIT |
Notable symptoms: Whole-board death within days-to-weeks; “red hash2”; −273°C / 120–180°C temp misreads.
Repair-cost signal: No dedicated public menu isolated — D-Central tiered (from $65 CAD diagnostic; L1/L2/L3 $95/$145/$195 CAD); third-party mail-in services exist.
Used-market signal: Used around US$438–$550 (three listings: 85T $505, 93T $550, 90T $438) ≈ CAD $611–$767. An indicative range, not a median.
Accuracy note: Chip-level TH per chip is cross-checked against the deeper Avalon reference before being treated as canonical.
Avalon A1366 A3210 family Grade C · Limited
Verified spec: shares the A1366/1346/1446/1466 control board and A32xx tooling.
Aggregated from 1 community-reported component-failure family across REDDIT, YT, RWEB. Data strength: THIN.
| Component | Failure mode (reported) | Frequency | Sources |
|---|---|---|---|
| Hashboard (ASIC + solder) | Same chip-dropout family as A1246; shares 1366/1346/1446/1466 control board + A32xx tooling (no isolated used price; flag honestly) | reported | 3 REDDIT · YT · RWEB |
Notable symptoms: Same chip-dropout family as the A1246.
Repair-cost signal: No dedicated public menu isolated — served under D-Central Avalon tiered.
Used-market signal: None isolated (parts / repair support confirms installed base only).
Accuracy note: Thin — graded honestly.
Avalon A14 (A1466) / A15 A3210 / A3206 Grade C · Limited
Verified spec: newer Avalon generation; lighter North-American repair-menu and used-price coverage.
Aggregated from 3 community-reported component-failure families across YT, RWEB. Data strength: THIN-MODERATE.
| Component | Failure mode (reported) | Frequency | Sources |
|---|---|---|---|
| Hashboard (ASIC + solder) | Dead chip - firmware skips failed chip, headline hashrate drops proportionally | frequently reported | 2 YT · RWEB |
| Other (cabling, connectors, sensors, firmware, thermal interface) | Thermal-paste dry-out (9-12 mo A15); dust ('30 days = 5-12% drop' vendor estimate, attribute); voltage-domain caps/PMIC (PVT_V scatter >+-30mV); AC <220V derating; connector oxidation (keep website copy per-DOMAIN for regulation (VERIFY V14)) | frequently to occasionally reported | 1 RWEB |
| Other (cabling, connectors, sensors, firmware, thermal interface) | HW-error% spikes (marginal pool / OTA regression / degrading chip) | occasionally reported | 1 YT |
Notable symptoms: Dead chip — firmware skips the failed chip and headline hashrate drops proportionally; HW-error% spikes.
Repair-cost signal: The A1566 service is listed up to ~$1,100 CAD; D-Central services through the A1466 and beyond; replacement hashboards are sold.
Used-market signal: A15 new around US$2,499 (215 TH) ≈ CAD $3,486; A15XP ~US$3,849 ≈ CAD $5,369. A14 and A15 used = none isolated.
Accuracy note: Avalon exposes per-chip PVT_V telemetry (a readout), but voltage regulation remains per-DOMAIN — per-chip readout is normal, per-chip control is not.
Open-source boards (Bitaxe / NerdQaxe / NerdOctaxe / NerdAxe)
Bitaxe Gamma BM1370 Grade A · Strong
Verified spec: ~1.0–1.5 TH/s single-chip board, ~16–18 W; per-chip baseline ~750+ GH/s; 5 V barrel jack
Aggregated from 5 community-reported component-failure families across REDDIT, YT, RCOST, RWEB, OSS. Data strength: STRONG.
| Component | Failure mode (reported) | Frequency | Sources |
|---|---|---|---|
| Power domain (LDO / caps / boost / VRM) | TPS546 (TPS546D24S) VCORE regulator - narrow VIN 4.8-5.3V; sag <4.5V trips 'Power Fault Detected'; can latch needing physical disconnect; 105C VRM thermal shutdown. Worst on Gamma (VRM; VERIFY VIN window V8 / part V7) | frequently reported | 5 REDDIT · YT · RCOST · RWEB · OSS |
| Hashboard (ASIC + solder) | BM1370 '0 GH/s pulling ~5W' - soft/API restart does NOT recover; only full cold power cycle clears it. GitHub #588 = '>=3 of reporter units' (single multi-device anecdote, NOT a rate) | frequently reported (Gamma-specific) | 5 REDDIT · YT · RCOST · RWEB · OSS |
| Power domain (LDO / caps / boost / VRM) | VR/VRM thermal bottleneck (not the ASIC) - VR warning ~70C, 'pops' sustained >80C; AxeOS shows only ASIC temp on OLED -> users cook the VR first | frequently reported | 2 OSS · RWEB |
| Other (cabling, connectors, sensors, firmware, thermal interface) | TPS546 boot-race ('i2c handle not initialized' / Device ID ff:ff...) - firmware timing regression 2.8.0+ (wontfix); FIRMWARE not a failed part (GitHub #1291; firmware) | occasionally reported (subset) | 1 OSS |
| Other (cabling, connectors, sensors, firmware, thermal interface) | Thermal paste / heatsink gap (shipping vibration); power MOSFETs/DC-DC; ESP32-S3 firmware corruption / 2.4GHz-only WiFi; AxeOS v2.13/2.14 telemetry regression (thermal/firmware/WiFi) | frequently to occasionally reported | 4 REDDIT · YT · RWEB · OSS |
Notable symptoms: “Power Fault Detected”; 0 GH/s at ~5 W (vs ~16–17 W); metrics read ‘--’/null while hashing; >75°C auto-throttle.
Repair-cost signal: Single-board repair is listed around US$50 base (4–5 day, 30-day warranty) by third-party shops; D-Central ASIC repair from $65 CAD. The VCORE regulator (TPS546D24S) is a low-cost bench part.
Used-market signal: New retail only. Gamma + PSU from ~CAD $171 on sale; Gamma++ and turbo/“Ice” variants higher. Thin used market.
Accuracy note: Bitaxe Gamma = BM1370 is correct and preserved (the isolated catalog row that says otherwise is a known error). The 0-GH/s-needs-cold-boot behaviour is BM1370-specific; a GitHub report covering several of one reporter's units is a single multi-device anecdote, not a rate.
Bitaxe Supra BM1368 Grade B · Moderate
Verified spec: single-chip board; 5 V barrel jack
Aggregated from 4 community-reported component-failure families across REDDIT, YT, RWEB, OSS. Data strength: MODERATE-STRONG.
| Component | Failure mode (reported) | Frequency | Sources |
|---|---|---|---|
| Power domain (LDO / caps / boost / VRM) | TPS546 VRM - 'Power Fault Detected' when output leaves regulation; cheap '5V' PSUs sag to 4.5V under load (more tolerant than Gamma; API restart recovers) | frequently reported | 3 REDDIT · YT · RWEB |
| Hashboard (ASIC + solder) | BM1368 BGA joint - ~50% hashrate, elevated HW-error% (silent throttle) -> reflow | frequently reported | 2 YT · RWEB |
| Fan / cooling | Stock fan 'crap' (README) - mandatory active cooling, Noctua NF-A4x20 swap | frequently reported | 1 OSS |
| Other (cabling, connectors, sensors, firmware, thermal interface) | Barrel jack (5.5x2.1mm) works loose; I2C sensor subsystem reads null (jack/sensor) | occasionally reported | 1 REDDIT |
Notable symptoms: “Power Fault Detected” when output leaves regulation; ~50% hashrate with elevated HW-error% (silent throttle) → reflow.
Repair-cost signal: Single-board repair around US$50 (Bitaxe shop class); replacement BM1368 chips are stocked.
Used-market signal: New only. Supra + PSU from ~CAD $174; SupraHex higher.
Accuracy note: More tolerant than the Gamma — the BM1370 bring-up faults do not affect the BM1368 the same way (an API restart recovers).
Bitaxe Ultra BM1366 Grade B · Moderate
Verified spec: single-chip board; 5 V barrel jack
Aggregated from 2 community-reported component-failure families across OSS, RCOST, REDDIT. Data strength: MODERATE.
| Component | Failure mode (reported) | Frequency | Sources |
|---|---|---|---|
| Fan / cooling | EMC2101 PWM fan controller - cooling-control IC; no notable Ultra-specific hardware errata; 'most battle-tested Bitaxe chip', recovers on soft restart (deliberately low errata = the finding) | single anecdotal report / low | 1 OSS |
| Power domain (LDO / caps / boost / VRM) | TPS546 VRM (shared family); zero-hashrate after power cycle - most issues external (PSU/pool), not chip | occasionally reported | 2 RCOST · REDDIT |
Notable symptoms: Zero hashrate after a power cycle — most issues are external (PSU / pool), not the chip.
Repair-cost signal: Single-board repair around US$50 (Bitaxe shop class).
Used-market signal: New only. Ultra + PSU around CAD $160.
Accuracy note: Deliberately low errata — that is the finding. The most battle-tested Bitaxe chip; recovers on a soft restart.
NerdQaxe+ / NerdQaxe++ 4× BM1368 (NerdQaxe+) / 4× BM1370 (NerdQaxe++) Grade A · Strong
Verified spec: four-chip daisy chain; 12 V XT30 input
Aggregated from 5 community-reported component-failure families across REDDIT, YT, RCOST, RWEB, OSS. Data strength: STRONG.
| Component | Failure mode (reported) | Frequency | Sources |
|---|---|---|---|
| Power domain (LDO / caps / boost / VRM) | VRM CSD95472Q5MC blown -> 12V-rail dead short (0-2ohm continuity) - 'VRMs, not ASICs, are usually the culprit' when OC'd; blows onboard SMD fuse. DOMINANT overclock death mode | frequently reported | 4 REDDIT · YT · RCOST · RWEB |
| PSU | Hidden 5x20mm glass mainboard fuse (250V 5A fast-blow) - won't power on (no fan/display) but power light still on (BITSOLOPLAYER; fuse) | frequently reported | 1 YT |
| Other (cabling, connectors, sensors, firmware, thermal interface) | Thermal paste / heatsink gap -> PSU error #0000015 - shipping vibration opens gap; reapply MX-4/6W+ (thermal interface) | frequently reported | 4 REDDIT · YT · RCOST · RWEB |
| Hashboard (ASIC + solder) | Dead chip in 4-chip daisy chain = ~25% hashrate loss; loose XT30 -> intermittent power/chip-detect; real draw ~80W vs ~60W 'rated'; misaligned pins -> 3.3V short (connector/hashboard) | occasionally to frequently reported | 3 OSS · REDDIT · YT |
| Other (cabling, connectors, sensors, firmware, thermal interface) | 2.4GHz-only WiFi 'No AP Found'; random Software Panic resets (firmware) (WiFi/firmware; VERIFY fuse rating V9) | frequently to occasionally reported | 1 OSS |
Notable symptoms: Dead board, no power; PSU error #0000015; 0–2Ω across the power connector; −25% hashrate from one dead chip.
Repair-cost signal: Largely community / DIY — no shop figure; consumables only (a sub-$1 SMD fuse, ~$8–12 thermal paste).
Used-market signal: New only. NerdQaxe+ ~CAD-equivalent of US$379.99; NerdQaxe++ from ~US$299.99 on sale; hydro variants higher.
Accuracy note: On overclocked units the VRM (CSD95472Q5MC), not the ASIC, is usually the culprit — it blows the onboard SMD fuse. Real draw (~80 W) exceeds the ~60 W “rated” figure.
NerdOctaxe Gamma 8× BM1370 Grade B · Moderate
Verified spec: eight-chip board; 12 V XT60 ≥18–20 A; ~160–180 W; ~9.6 TH
Aggregated from 3 community-reported component-failure families across REDDIT, YT, OSS. Data strength: MODERATE.
| Component | Failure mode (reported) | Frequency | Sources |
|---|---|---|---|
| Other (cabling, connectors, sensors, firmware, thermal interface) | TPS53647 multiphase current-protection -> 'Guru Meditation #51A00000' boot crash on fw <=v1.0.35 (140A@1.15V ~= 161W ceiling exceeded) - fixed v1.0.36 (160A). FIRMWARE-config bug (firmware/power-protection) | frequently reported (affected fw) | 3 REDDIT · YT · OSS |
| Fan / cooling | Disconnected/half-seated cooler fan -> 4 of 8 chips overheat in minutes; half-seated XT60 overheats under ~20A; copper-weight/fuse variance between vendor builds (1/1oz vs 2/2oz) (fan/connector) | occasionally reported / single anecdotal | 1 OSS |
| Power domain (LDO / caps / boost / VRM) | VR / hashboard cooling failures (hand-assembled) | occasionally reported | 1 YT |
Notable symptoms: “Guru Meditation #51A00000” boot crash on firmware ≤v1.0.35 (current-protection ceiling) — a firmware-config bug fixed in v1.0.36.
Repair-cost signal: Community / DIY — no shop figure isolated.
Used-market signal: New only. Around US$649.99 / €699 (9.6 TH); hydro higher.
Accuracy note: The headline boot crash is a firmware-configuration issue, not a failed part. A half-seated cooler fan can overheat four of eight chips within minutes.
NerdAxe BM1370 (Gamma) / BM1366 (Ultra) Grade C · Limited
Verified spec: single-chip fork of the Bitaxe; 5 V / TPS546 family
Aggregated from 2 community-reported component-failure families across OSS. Data strength: THIN.
| Component | Failure mode (reported) | Frequency | Sources |
|---|---|---|---|
| Fan / cooling | EMC2101 fan PWM scaling bug (50% setting -> ~70% RPM) - fixed in fork PR #2 (+ signed-temp register fix). FIRMWARE not a part (firmware) | occasionally reported | 1 OSS |
| Power domain (LDO / caps / boost / VRM) | Shares Bitaxe 5V/TPS546/single-ASIC family; wrong-target firmware (I2C pins 43/44 vs Bitaxe 47/48) can make a board appear 'dead' (mostly shared-family + firmware; flag honestly) | reported | 1 OSS |
Notable symptoms: A fan-PWM scaling bug (50% setting → ~70% RPM) — a firmware issue fixed in the fork.
Repair-cost signal: Community / DIY.
Used-market signal: New only. NerdAxe + PSU around US$199.99 ≈ CAD $279.
Accuracy note: Mostly shared-family and firmware issues. Wrong-target firmware (different I2C pin mapping) can make a board look “dead” when it is not. Graded honestly.
Power supplies (PSUs)
Bitmain APW3 / APW3++ PSU Grade A · Strong
Verified spec: S7/S9-era PSU. 11.6–13.0 V, 133 A, 1600 W @220 V / 1200 W @110 V
Aggregated from 4 community-reported component-failure families across REDDIT, RWEB. Data strength: STRONG.
| Component | Failure mode (reported) | Frequency | Sources |
|---|---|---|---|
| PSU | Cold-start failure <10C (fanatic26 'notorious if ambient 10c or under'; ccgllc <1/3 of 80 booted near-freezing). Remedy: pre-warm (S7/S9-era) | frequently reported | 1 REDDIT |
| PSU | Residual-charge lockout - must wait for fan to fully stop before re-power or it never outputs 12V (sequencing) | frequently reported | 1 REDDIT |
| PSU | Filter-cap degradation after 3-4 yr - high AC ripple; works light-load, fails under 3-board load (caps) | frequently reported | 2 REDDIT · RWEB |
| PSU | No/low DC output (fan still spins) - mshome: 0.347V@220V where 12V expected (output) | single anecdotal report (symptom class common) | 1 REDDIT |
Notable symptoms: Won't start below ~10°C ambient; residual-charge lockout if re-powered before the fan stops; high AC ripple after 3–4 years; 0.347 V where 12 V is expected.
Repair-cost signal: Often “not worth repairing” versus replacing (a single old anecdote); D-Central PSU repair $100–$300 CAD. A new replacement unit can be as low as ~US$14.
Used-market signal: Replacement units are inexpensive and widely available.
Accuracy note: Cold-start failure and residual-charge lockout are the signature, well-quoted behaviours.
Bitmain APW7 PSU Grade C · Limited
Verified spec: S9/L3+/D3-era PSU. 150 A, 1800 W @220 V
Aggregated from 1 community-reported component-failure family. Data strength: THIN.
| Component | Failure mode (reported) | Frequency | Sources |
|---|---|---|---|
| PSU | Folded into APW-series shop service; no distinct web failure or repair price isolated (S9/L3+/D3-era) | honest gap - no public data | 0 |
Notable symptoms: Folded into the APW-series shop service.
Repair-cost signal: Under the APW-series menu; D-Central $100–$300 CAD.
Accuracy note: No distinct public failure or repair price isolated — honest gap.
Bitmain APW9 / APW9+ PSU Grade B · Moderate
Verified spec: S17/T17-era PSU. 14.5–21 V, 170–200 A, 3600 W @220 V
Aggregated from 1 community-reported component-failure family across YT, REDDIT. Data strength: MODERATE.
| Component | Failure mode (reported) | Frequency | Sources |
|---|---|---|---|
| PSU | Output-stage caps fail under load; dead PSU / below-spec voltage before complete failure. Diagnostic: 14.0-15.0V no-load; fan never spins on AC -> internal failure (S17/T17-era; more reliable than APW3 but still fails) | occasionally to frequently reported | 2 YT · REDDIT |
Notable symptoms: Output-stage caps fail under load; dead PSU or below-spec voltage before complete failure. 14.0–15.0 V no-load; if the fan never spins on AC, internal failure.
Repair-cost signal: Under the APW-series menu; D-Central $100–$300 CAD.
Accuracy note: More reliable than the APW3 but still fails under load.
Bitmain APW12 PSU Grade A · Strong
Verified spec: S19-era PSU. 12–15 V, 233 A, 3600 W
Aggregated from 2 community-reported component-failure families across YT, RWEB, REDDIT. Data strength: STRONG.
| Component | Failure mode (reported) | Frequency | Sources |
|---|---|---|---|
| PSU | Output stage / caps -> voltage droop + intermittent reboots (clicking/buzzing/whining), can spike and damage boards (S19-era) | frequently reported | 3 YT · RWEB · REDDIT |
| PSU | 6-pin connector partial-seat -> heat / melt. Integration FACT: d/e/f add voltage feedback; a/b/c CANNOT replace d/e/f (connector) | frequently reported | 2 YT · RWEB |
Notable symptoms: Output-stage / cap failure → voltage droop and intermittent reboots (clicking / buzzing / whining); can spike and damage boards. 6-pin connector partial-seat → heat / melt.
Repair-cost signal: Board-level repair around US$200 (third-party); D-Central $100–$300 CAD. A new APW12-1215 replacement can be ~US$82 (150-day warranty).
Used-market signal: Replacement units widely available.
Accuracy note: Integration gotcha: d/e/f versions add voltage feedback — a/b/c units cannot replace d/e/f.
Bitmain APW17 PSU Grade C · Limited
Verified spec: S21-era PSU. 12–15 V, 267 A, 3600 W (S21 / S21 Pro / S21 XP / S19j XP / KS5)
Aggregated from 1 community-reported component-failure family across RCOST. Data strength: THIN.
| Component | Failure mode (reported) | Frequency | Sources |
|---|---|---|---|
| PSU | Sparse web failure data; APW171215 repair $200 USD (Antminer-Repair.com). Honest gap on failure modes (S21-era (S21/S21 Pro/S21 XP/S19j XP/KS5)) | honest gap - sparse public data | 0 RCOST |
Notable symptoms: Sparse public failure data.
Repair-cost signal: Board-level repair around US$200 (third-party).
Accuracy note: Honest gap on failure modes — graded honestly.
Whatsminer P21 / P221 / P222 PSU Grade A · Strong
Verified spec: P22-family WhatsPower PSU. 3300 W, 12 V-240 A + 12 V-12 A
Aggregated from 3 community-reported component-failure families across RWEB, REDDIT, RCOST. Data strength: STRONG.
| Component | Failure mode (reported) | Frequency | Sources |
|---|---|---|---|
| PSU | M6 copper-bar bolts loosen (ship ~2.0 N.m vs 2.5-3.5 spec) - AC LED green, fan spins, 0V or 0<->12V oscillation; LYS re-torque fixes '~95% of no-output cases' (vendor claim, attribute) (connector torque; 95% claim attributed not fact) | frequently reported | 2 RWEB · REDDIT |
| PSU | Primary MOSFETs (GP47S60X/IPA60R060P7/STW48N60DM2/OSG65R069HS/JCS12N65FT) - blown 20A fuse, blackening at Q ('high failure under surge') | frequently reported | 1 RWEB |
| PSU | HF32F-G relay; STM32F334R8T6 MCU; GBJ3510 bridge rectifier; NTMFS5C430NL sync MOSFET; NCP1399AC LLC; 500V 250uF bulk cap (dry-out >3yr); NTC 8D-20; 20A glass fuse (8-11V instead of 12V = sync MOSFET; ripple >0.1V = bulk cap) | occasionally reported / single anecdotal (MCU) | 2 RWEB · RCOST |
Notable symptoms: AC LED green + 0 V out; 8–11 V instead of 12 V (sync MOSFET); output ripple >0.1 V (bulk cap).
Repair-cost signal: No North-American shop price isolated; a spare-parts kit repairs roughly 3–5 PSUs per vendor estimate (wider “30–50 PSUs” is a marketing claim).
Used-market signal: Replacement units are sold but prices are not cleanly isolated — do not invent.
Accuracy note: A re-torque-fixes-“~95% of no-output cases” figure is a vendor claim, quoted with attribution.
Cross-model failure patterns
- Daisy-chain death — one dead or cold-solder chip disables the whole hashboard (downstream chips vanish). Seen across the S19, S19 Pro, S19j Pro, S21, T21, S17, L7 and Avalon families, and behind Whatsminer Error 542. The single most-reported ASIC failure family, explained by the series-wired hashboard architecture.
- Thermal-shock solder cracking on power-cycle — hot-to-cold cycling cracks marginal joints (“it was fine, then I rebooted it and a board died”). The 2025 headline finding on the S21/T21 generation.
- Power-domain failure — LDO drift plus 0.8 V/1.8 V filter-cap shorts and boost faults are the leading board-level “0 chips” cause across the S19, S19 XP, L7 and S21 families.
- Connectivity-first “not detected” — ribbon / signal cables, connectors, loose copper-bar bolts (Whatsminer) and controller-connector corrosion (Avalon) are the cheapest and most common not-detected causes.
- PSU is the dominant non-chip family — APW3++ cold-start and cap degradation; Whatsminer power-protection errors and copper-bolt torque; Bitaxe / NerdQaxe VRM and input-voltage sag.
- BM1370 voltage sensitivity — a narrow input window and a 0-GH/s-needs-cold-boot behaviour are documented in the Bitaxe Gamma ecosystem; extending this to the full S21 Pro/XP line is inference, flagged as such.
- VR/VRM is the open-source thermal bottleneck, not the ASIC — the regulator (CSD95472Q5MC on NerdQaxe, TPS546 on Bitaxe) overheats first, and the default telemetry often hides its temperature.
- Thermal-interface degradation — paste “measured in months”: S17 heatsink delamination (signature), Avalon dry-out, Bitaxe / NerdQaxe thermal-interface gaps.
- The S17/T17 line is the most-cited poor-reliability generation — a mix of community opinion and a third-party “30% failure” estimate (attributed, never presented as a measured rate).
- Newest flagships are sparse — S21 Pro/XP/+, Avalon A14/A15, Whatsminer M53/M56S are mostly in-warranty or commercial, so community data is thin. Stated honestly, not padded.
Cite this dataset
This dataset is free to use, share and adapt under the Creative Commons Attribution 4.0 International (CC BY 4.0) licence — please credit D-Central and link back.
Suggested attribution:
D-Central Technologies. "ASIC Reliability / Failure-Mode Index." d-central.tech, updated 2026-06-09. Licensed under CC BY 4.0. Accessed 13 June 2026. https://d-central.tech/data/asic-reliability-index/Related products, repair, and setup paths
- how D-Central diagnoses ASIC repairs
- ASIC troubleshooting library
- ASIC manuals and repair guides
- replacement hashboards
- ASIC control boards
- ASIC power supplies
- S19 family replacement hashboard
- C52 replacement control board
- APW12 S19 power supply
- immersion cooling hub
- home immersion cooling guide
- ASIC miners for immersion planning
- ASIC cooling parts
- airflow shroud before immersion
- compare miner specs in the database
- ASIC repair support
- compare ASIC miner specs
- ASIC miner database
- Antminer S19 specs and profitability
- buy a tested Antminer S19
- Antminer S19 maintenance guide
- Antminer S19 repair service
- Antminer S21 specs
- Bitmain Antminer S21
- Antminer S21 maintenance guide
- BM1370BC S21 Pro chip
- Antminer S9 specs
- Bitmain Antminer S9
- Antminer S9 maintenance guide
- S9 hashboard repair parts bundle
- Whatsminer M30S specs
- Whatsminer repair guide
- MicroBT Whatsminer M30S++
- Whatsminer M3x exhaust shroud
Last reviewed June 12, 2026.